Documentation of GEATbxReleaseNotes

Global Index (all files) (short | long) | Local contents | Local Index (files in subdir) (short | long)

Help text

 Release Notes of the GEATbx

 Notes about changes in the source code and the documentation of the 
 Genetic and Evolutionary Algorithm Toolbox for use with Matlab.

 See also: documentation of the GEATbx

 Author:   Hartmut Pohlheim
 GEATbx 3.8 - December 2006 (public version)
  - Renamed function for adding paths to search path (path_gea, was toolboxpath).
      The function for extending the Matlab search path was renamed to
      path_gea (old name was toolboxpath). Now the name is more consistent
      with the functionality (and multiple projects can use this naming 
      The actual path extension is now done using addpath (and does not use 
      the low level function matlabpath).
      The help inside the function was partly rewritten and extended with 
      another example. 
  - New demo function to demonstrate the optimization of integer variables.
  - Elegant methode for the handling of constraints employing additional constraints.
      Constraints are defined as additional objectives. If a constraint is
      not violated, the corresponding objective is zero. If a constraint is
      violated, die objective is identical to the violation. Additionally,
      the goals for the "constrained objectives" must be defined as 0 (zero).
      Thus, the "constrained objectives" are only optimized as long as the 
      constraint is violated. If no constraint is violated, only the main
      (functional) objectives are active.
      This elegant and universal constraint handling method is already 
      implemented in most of the constrained example functions.
      DEBs constrained function (mobjdebconstr)
      BELEGUNDUs function (mobjbelegundu)
      KITAs function (mobjkita)
      BAHINs function (mobjbahinconstr)
      SOLANDs function (mobjsoland)
  - Added the definition of Goals inside multi-objective functions.
      It is now possible to define standard values for goals inside the 
      (multi)objective functions.
      To get the defined goals use: geaobjpara('NameofObjFun',  [16])
      This is similar to defining the upper and lower bounds of the 
      variables (VLUB/'VarBoundMin'/'VarBoundMax') or the default number 
      of variables ('NumVarDefault').
      The respective m-files geaobjpara and objfunoptset are extended.
      An example can be found in the demo function demomop.
      Most of the multi-objective example functions use the new method.
  - Added further examples to demo function demomop for multi-objective optimization.
      BELEGUNDUs function (mobjbelegundu)
      KITAs function (mobjkita)
  - Extension of multi-objective demo function demomop.
      Added assessment of best individuals at the end of a multi-objective 
      optimization (objective value calculation, fitness values, plot of 
      PARETO front).
      Now, the PARETO front of the best solutions found during optimization 
      is visualized. Additionally, the fitness (FitnV) and rank values (RankMOV)
      of the best solutions are available in the calling workspace and can be used
      for further calculations.
  - Removed a bug in colbestind, appearing with binary values.
      Added an abs before the sum of diff calculation.
      Examples were added for better understanding (with binary population)
      of the functionality.
  - New/separate Section in Introduction to Evolutionary Algorithms about:
      References for publications regarding multi-objective optimization.
  - New Chapter on multi-objective optimization in Introduction:
      Moved PARETO-ranking, goal attainment, sharing and weighted sum 
      to the new chapter.
  - New Chapter on the use of Multi-objective optimization in Tutorial:
      Comprehensive explanation of how to switch on MO optimization, the 
      parameters to adjust, the provided visualization and the demo function
  - New Chapter on Constrained optimization in Tutorial:
      Comprehensive explanation of how to use constraint optimization. 
      Variable constraints and functional constraint are explained and 
      the use demonstrated.
  - Section on Variable Representation in Tutorial considerably extended
      The aspets of different representations are described including the 
      conversion between binary to integer or real representation. The 
      examples were extended and a few more added including references to 
      the repective demo functions.
      Added information on the handling of different representations in 
      one optimization. Integer and binary representation can be used 
      simultaneously today. Real and binary or integer will be available 
      in a future version.
 GEATbx 3.7 - November 2005 (public version)
  - Removed a bug in colbestind.m, which prevented the correct collection of
      individuals when comparing the variables (sum for diff was operating along 
      the rows and not along the columns). Thus, for integer populations only a 
      few individuals were collected.
  - Added VRP (vehicle routing problem) example(s)
      Vehicle routing is a large class of ordering/permutation problems. To support 
      the start into this optimization area I wrote a demo function (demovrp), an 
      objective function (objvrp), a VRP specific plot function (plotvrp) and a few 
      utility functions (vrpconvind, vrpgetdata). Two data sets (vrpnc1, vrpsolrc101)
      are directly available and can be used for your first experiments. 
      (The vrp example needs routines from the matlog toolbox. Please get 
       Matlog and install it in your Matlab serach path.)
  - Help section updated for:
      demofun1cmp: show competition during optimization
  - Reworked all Contents.m files of the GEATbx
      The files are listed in named section, which give a much better overview
      on the available m-files. Functions belonging to one topic are grouped 
      In the M-Functions index of the documentation the Contents files are
      much more useful now.
  - Inclusion of objective function example documentation
      The inclusion of better graphics of the search space of many of the example 
      functions provides a much better understanding of the properties of the
      functions (all produced with plotmesh using the high level script graf_msh).
  - Index over all written documentation extended
      One index covering all parts of the written documentation is available. It
      covers the Introduction, the Tutorial, the Example function documentation 
      and the Options documentation.
  - New Chapter in Introduction about:
      Combination of Operators and Options to Produce Evolutionary Algorithms
      Extensive discussion of the different aspects when combining operators and 
      their options.
  GEATbx 3.65 - May 2005 (internal version)
  - Inclusion of additional multi-objective example functions.
      Further Deb, Thiele, Zitzler, Laumanns functions (mobjdtlz2, mobjdtlz3)
      Kita's function (mobjkita)
      Belegundu's function (mobjbelegundu)
  - Initialization of integer population (initip), (automatic) extension of bounds
      The bounds (VLUB) for integer representation are (automatically) extended 
      by (nearly) 0.5 for uniform initialization at the defined bounds.
      Now it is ensured, that the values are created uniform at random, even at the 
      bounds of the initialization range. Formerly, the bound values were only used 
      half as often as the other values. 
      See help initip for a longer description and two examples, which demonstrate 
      this effect.
      For integer representation you must use integer values for the bounds. Otherwise 
      the (automatic) extension breaks your bounds and you get variable values outside
      your defined bounds.
  GEATbx 3.6 - March 2005 (internal version)
  - Inclusion of additional multi-objective example functions.
      First Deb, Thiele, Zitzler, Laumanns function (mobjdtlz1)
      Quagliarella's function (mobjquagliarella.m)
      Fonseca's functions (mobjfonseca1 and mobjfonseca2) are already included.
  - Added information on multi-objective optimization to the Introduction.
      Multi-objective optimization is already supported by the GEATbx. However, 
      the documentation was still lacking. Now a whole section is dedicated to 
      the different mechanisms necessary for multi-objective ranking, including 
      PARETO-ranking, goal attainment and sharing.
      See Introduction, Chapter Selection, Subsection Multi-objective ranking
      The implemented archive (CollectBestInd) will be described later.
  - Corrected an error in documentation of equation for non-linear ranking.
      There was a (very old) error in the (sub)equation for non-linear ranking.
      See Introduction, Chapter Selection, Subsection Rank-based fitness assignment
      equation "X is computed as the root of the polynomial:"
      correct: 0 = (SP - Nind) ...        wrong: 0 = (SP - 1) ...
      The source code in ranking.m was always correct.
  GEATbx 3.55 - November 2004 (internal version)
  - Inclusion of additional multi-objective example functions.
      Cantilever beam problem (mobjcantilever and democantilever)
  - Rewrite of the objective function examples started.
      All parametric functions (objfun1-objfun12 and a few more) were 
      rewritten in the GEATbx documentation format. This format allows 
      the creation of pdf and html documents from the same source 
      (already used for Introduction, Tutorial and Options).
  GEATbx 3.5 - February 2004 (public version)
  - Multi-objective ranking is used wherever necessary and only when necessary.
  - Sharing (by default sharing in search space) is carried out directly after ranking. 
      However, sharing is only carried out when needed (means before selection).
      Sharing in solution space can be selected as well. The switch to sharing in 
      solution space can be done at the end of ranking.m inside the source code 
      (see ShareMethod).
  - Three different implementations of MO ranking are included in
      ranking.m. Depending on the number of individuals and objectives (automatically)
      the most appropriate is chosen (in such a way, that not too much memory is used).
      The first method takes next to no time but a lot of memory and uses 3-D
      The second requires linear time and memory and can handle goals.
      The third takes only a very small amount of memory and can handle goals as 
      well, but employs double for loops.
      It is always tried to use the quickest algorithm, but switch back if
      the amount of memory needed is larger than 32 MByte(or 128MByte in 2005).
      Have a look inside ranking.m for more info.
  - geamain2 returns a whole population at the end of the population.
      In the MO case this population contains all non-dominated solutions
      found (but not more than individuals inside the population).
  - colbestind (CollectBest.*) handles MO individuals as well. These
      results are included in the return population as well.
  - Possible problems when changing from version 3.3 to version 3.5 (or newer):
      Some sub functions in compdiv/compdiv2 are now separate functions.
      If you use any of these functions in your projects please change the calls.
      To get the parameters of the objective function (boundaries of variables, 
      long name of function, global optimum ...) different interfaces were used 
      during the last years. Now one function (geaobjpara.m) provides one 
      interface to everything. Have a look at demomop and demogeatbx for an 
      example of the application.
  - Corrected error in calculation of objective tsp lib function.
      computing dist contained wrong reference (irout instead of i)
      correct: for irout = 1:Nroutes, dist(irout) = sum(diag ...
      wrong:   for irout = 1:Nroutes, dist(i)     = sum(diag ...
 End of Release Notes GEATbx (only 2004 and newer)
GEATbx: Main page  Tutorial  Algorithms  M-functions  Parameter/Options  Example functions 

This document is part of version 3.8 of the GEATbx: Genetic and Evolutionary Algorithm Toolbox for use with Matlab -
The Genetic and Evolutionary Algorithm Toolbox is not public domain.
© 1994-2006 Hartmut Pohlheim, All Rights Reserved, (